AI Research Brief
Archives
Search...
Log in
Subscribe
Safety Benchmark Score Ranges Differ 23-Fold
October 21, 2026
Safety Benchmark Scores May Reflect Policy Definitions More Than Capability. Ten guardrails shared the same architecture and optimization pipeline. Their...
Self-Conditioning Makes Extra Sampling Steps Matter
October 17, 2026
Extra Sampling Steps Don't Guarantee Deeper Reasoning. Under specific conditions, constant-round parallel computation can simulate standard flow-matching...
DoSA Delivers Up to a 1.94x Time-to-First-Token Speedup over SGLang
October 16, 2026
DoSA Safely Skips Candidates That Cannot Reach Top-K. Provable bounds avoid full scoring for rejected candidates, delivering up to 1.94x and 1.50x time-to-...
AesCode 8B Gains 31.1 Points in Its Own Benchmark
October 16, 2026
AesCode Separates Verifiable Correctness From Aesthetics. Its 8B model improved visual quality by 31.1 percentage points over a same-size Qwen3-VL baseline...
Multics Scales Four-Player Training to Ten-Player Inference
October 15, 2026
Hierarchical Communication Removes Full Connectivity. Multics routes information through player, group, and global layers. It scales directly from four-...
Atomic Interfaces Cut Desktop Agent Tokens by 73%
October 15, 2026
Better Interface Design Improves Desktop Agent Accuracy and Cost: Across eight applications and 368 tasks, atomic interfaces raised two models’ success rates...
Artisan Posts Lowest Collision Rate Across 10+ Representative Models
October 12, 2026
Artisan Brings 3D Assets Into Physical Validation. It recorded the lowest collision rate and highest stability rate among more than 10 representative models....
Token Routing Lifts Decode Throughput by 2.01–64.15x
October 10, 2026
Token-Level Routing Breaks Existing Batching Assumptions First: TokenRouter separates request-level routing logic from asynchronous per-model execution,...
Causal Pretraining Makes Long Robot Context Pay Off
October 9, 2026
Causal Pretraining Determines Whether Long Histories Improve Control: Extending video history from zero to 19.2 seconds raised RoboCasa GR-1 success from...
1T Cross-Region Relay-Tree Refit in 150 Seconds at 3% Change Rate
October 8, 2026
A 1T Relay-Tree Refit Takes 150 Seconds at a 3% Change Rate. The paper measured 87.5 minutes for a complete 1T checkpoint transfer between two AWS regions....
Agentic Retrieval Gains 8.7 nDCG@10 Points Over Standard Retrieval With the Same Embedding Model
October 7, 2026
Agentic Retrieval Trades 160× Latency for 8.7 nDCG@10 Points. With the same embedding model, average query latency rose from 0.67 to 107.4 seconds while...
Prism Trains 2.5× Faster and Improves Quality
October 6, 2026
Watermarks Change Generation Paths and Can Cause Factual Errors: All six tested methods showed this effect. At a 0.90 detection true positive rate and 1%...
Pivot-SD Wins With 200 Problems; Queen Gains 915 Elo
October 5, 2026
Pivot-SD Beats Full-Sequence SFT With Sparse Supervision. It used 200 problems and four samples each to outperform full-sequence SFT and same-budget...
Mean OSWorld Task Score Rises From 33.0% to 40.8%
October 3, 2026
Post-Training Can Improve One-Shot Accuracy While Shrinking Solution Coverage. Across 42 comparisons spanning four model families and three agent benchmarks,...
Chunked KV Compression Shows Retrieval Gaps of Up to 40 Percentage Points
October 2, 2026
KV Compression Windows Create Periodic Retrieval Weaknesses: The same information can vary by up to 40 percentage points in accuracy across phases....
TRM Defines Rubrics; SAKI Reaches 4.22× Throughput
October 2, 2026
TRM Defines Rubrics Before Assigning Fine-Grained Scores. PD-GRPO reduces the score polarization caused by conventional pairwise preference optimization,...
Across 600 Hard Cases, RSB Searches 50% Fewer Branches Than State-of-the-Art Heuristics
September 30, 2026
RL Helps Verifiers Avoid Dead Ends. RSB reranks branching heuristics with reinforcement learning. Across 600 hard instances, it solves 11% more while...
250K Deployment Traces Become Tests, VQS Answers Reach 94%
September 29, 2026
Production Failures Can Become Regression Tests: TraceDance turns 252,557 real agent sessions into 107 targeted benchmarks. Both annotators confirmed the...
Qwen3.5-4B’s Nine-Benchmark Average Retention Reaches 82.3% at 5% Visual Tokens
September 29, 2026
Relevant Documents Are Not a Good Evidence Set. AdaTutoRank adjusts tutoring intensity to candidate-set quality and distills guidance gains into per-token...
AgentWorld's Best Model Succeeds Only 52% of the Time
September 28, 2026
Task Success Does Not Prove Good Multi-Agent Collaboration: AgentWorld tests 3–20 agents with asymmetric roles across 100 human-annotated tasks. The best...
Older archives